SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/btrfs/file.c

   1 /*
   2  * Copyright (C) 2007 Oracle.  All rights reserved.
   3  *
   4  * This program is free software; you can redistribute it and/or
   5  * modify it under the terms of the GNU General Public
   6  * License v2 as published by the Free Software Foundation.
   7  *
   8  * This program is distributed in the hope that it will be useful,
   9  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  10  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  11  * General Public License for more details.
  12  *
  13  * You should have received a copy of the GNU General Public
  14  * License along with this program; if not, write to the
  15  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
  16  * Boston, MA 021110-1307, USA.
  17  */
  18
  19 #include <linux/fs.h>
  20 #include <linux/pagemap.h>
  21 #include <linux/highmem.h>
  22 #include <linux/time.h>
  23 #include <linux/init.h>
  24 #include <linux/string.h>
  25 #include <linux/backing-dev.h>
  26 #include <linux/mpage.h>
  27 #include <linux/swap.h>
  28 #include <linux/writeback.h>
  29 #include <linux/statfs.h>
  30 #include <linux/compat.h>
  31 #include "ctree.h"
  32 #include "disk-io.h"
  33 #include "transaction.h"
  34 #include "btrfs_inode.h"
  35 #include "ioctl.h"
  36 #include "print-tree.h"
  37 #include "tree-log.h"
  38 #include "locking.h"
  39 #include "compat.h"
  40
  41
  42 /* simple helper to fault in pages and copy.  This should go away
  43  * and be replaced with calls into generic code.
  44  */
  45 static noinline int btrfs_copy_from_user(loff_t pos, int num_pages,
  46                                          int write_bytes,
  47                                          struct page **prepared_pages,
  48                                          const char __user *buf)
  49 {
  50         long page_fault = 0;
  51         int i;
  52         int offset = pos & (PAGE_CACHE_SIZE - 1);
  53
  54         for (i = 0; i < num_pages && write_bytes > 0; i++, offset = 0) {
  55                 size_t count = min_t(size_t,
  56                                      PAGE_CACHE_SIZE - offset, write_bytes);
  57                 struct page *page = prepared_pages[i];
  58                 fault_in_pages_readable(buf, count);
  59
  60                 /* Copy data from userspace to the current page */
  61                 kmap(page);
  62                 page_fault = __copy_from_user(page_address(page) + offset,
  63                                               buf, count);
  64                 /* Flush processor's dcache for this page */
  65                 flush_dcache_page(page);
  66                 kunmap(page);
  67                 buf += count;
  68                 write_bytes -= count;
  69
  70                 if (page_fault)
  71                         break;
  72         }
  73         return page_fault ? -EFAULT : 0;
  74 }
  75
  76 /*
  77  * unlocks pages after btrfs_file_write is done with them
  78  */
  79 static noinline void btrfs_drop_pages(struct page **pages, size_t num_pages)
  80 {
  81         size_t i;
  82         for (i = 0; i < num_pages; i++) {
  83                 if (!pages[i])
  84                         break;
  85                 /* page checked is some magic around finding pages that
  86                  * have been modified without going through btrfs_set_page_dirty
  87                  * clear it here
  88                  */
  89                 ClearPageChecked(pages[i]);
  90                 unlock_page(pages[i]);
  91                 mark_page_accessed(pages[i]);
  92                 page_cache_release(pages[i]);
  93         }
  94 }
  95
  96 /*
  97  * after copy_from_user, pages need to be dirtied and we need to make
  98  * sure holes are created between the current EOF and the start of
  99  * any next extents (if required).
 100  *
 101  * this also makes the decision about creating an inline extent vs
 102  * doing real data extents, marking pages dirty and delalloc as required.
 103  */
 104 static noinline int dirty_and_release_pages(struct btrfs_trans_handle *trans,
 105                                    struct btrfs_root *root,
 106                                    struct file *file,
 107                                    struct page **pages,
 108                                    size_t num_pages,
 109                                    loff_t pos,
 110                                    size_t write_bytes)
 111 {
 112         int err = 0;
 113         int i;
 114         struct inode *inode = fdentry(file)->d_inode;
 115         u64 num_bytes;
 116         u64 start_pos;
 117         u64 end_of_last_block;
 118         u64 end_pos = pos + write_bytes;
 119         loff_t isize = i_size_read(inode);
 120
 121         start_pos = pos & ~((u64)root->sectorsize - 1);
 122         num_bytes = (write_bytes + pos - start_pos +
 123                     root->sectorsize - 1) & ~((u64)root->sectorsize - 1);
 124
 125         end_of_last_block = start_pos + num_bytes - 1;
 126         err = btrfs_set_extent_delalloc(inode, start_pos, end_of_last_block);
 127         if (err)
 128                 return err;
 129
 130         for (i = 0; i < num_pages; i++) {
 131                 struct page *p = pages[i];
 132                 SetPageUptodate(p);
 133                 ClearPageChecked(p);
 134                 set_page_dirty(p);
 135         }
 136         if (end_pos > isize) {
 137                 i_size_write(inode, end_pos);
 138                 /* we've only changed i_size in ram, and we haven't updated
 139                  * the disk i_size.  There is no need to log the inode
 140                  * at this time.
 141                  */
 142         }
 143         return err;
 144 }
 145
 146 /*
 147  * this drops all the extents in the cache that intersect the range
 148  * [start, end].  Existing extents are split as required.
 149  */
 150 int btrfs_drop_extent_cache(struct inode *inode, u64 start, u64 end,
 151                             int skip_pinned)
 152 {
 153         struct extent_map *em;
 154         struct extent_map *split = NULL;
 155         struct extent_map *split2 = NULL;
 156         struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree;
 157         u64 len = end - start + 1;
 158         int ret;
 159         int testend = 1;
 160         unsigned long flags;
 161         int compressed = 0;
 162
 163         WARN_ON(end < start);
 164         if (end == (u64)-1) {
 165                 len = (u64)-1;
 166                 testend = 0;
 167         }
 168         while (1) {
 169                 if (!split)
 170                         split = alloc_extent_map(GFP_NOFS);
 171                 if (!split2)
 172                         split2 = alloc_extent_map(GFP_NOFS);
 173
 174                 write_lock(&em_tree->lock);
 175                 em = lookup_extent_mapping(em_tree, start, len);
 176                 if (!em) {
 177                         write_unlock(&em_tree->lock);
 178                         break;
 179                 }
 180                 flags = em->flags;
 181                 if (skip_pinned && test_bit(EXTENT_FLAG_PINNED, &em->flags)) {
 182                         if (em->start <= start &&
 183                             (!testend || em->start + em->len >= start + len)) {
 184                                 free_extent_map(em);
 185                                 write_unlock(&em_tree->lock);
 186                                 break;
 187                         }
 188                         if (start < em->start) {
 189                                 len = em->start - start;
 190                         } else {
 191                                 len = start + len - (em->start + em->len);
 192                                 start = em->start + em->len;
 193                         }
 194                         free_extent_map(em);
 195                         write_unlock(&em_tree->lock);
 196                         continue;
 197                 }
 198                 compressed = test_bit(EXTENT_FLAG_COMPRESSED, &em->flags);
 199                 clear_bit(EXTENT_FLAG_PINNED, &em->flags);
 200                 remove_extent_mapping(em_tree, em);
 201
 202                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 203                     em->start < start) {
 204                         split->start = em->start;
 205                         split->len = start - em->start;
 206                         split->orig_start = em->orig_start;
 207                         split->block_start = em->block_start;
 208
 209                         if (compressed)
 210                                 split->block_len = em->block_len;
 211                         else
 212                                 split->block_len = split->len;
 213
 214                         split->bdev = em->bdev;
 215                         split->flags = flags;
 216                         ret = add_extent_mapping(em_tree, split);
 217                         BUG_ON(ret);
 218                         free_extent_map(split);
 219                         split = split2;
 220                         split2 = NULL;
 221                 }
 222                 if (em->block_start < EXTENT_MAP_LAST_BYTE &&
 223                     testend && em->start + em->len > start + len) {
 224                         u64 diff = start + len - em->start;
 225
 226                         split->start = start + len;
 227                         split->len = em->start + em->len - (start + len);
 228                         split->bdev = em->bdev;
 229                         split->flags = flags;
 230
 231                         if (compressed) {
 232                                 split->block_len = em->block_len;
 233                                 split->block_start = em->block_start;
 234                                 split->orig_start = em->orig_start;
 235                         } else {
 236                                 split->block_len = split->len;
 237                                 split->block_start = em->block_start + diff;
 238                                 split->orig_start = split->start;
 239                         }
 240
 241                         ret = add_extent_mapping(em_tree, split);
 242                         BUG_ON(ret);
 243                         free_extent_map(split);
 244                         split = NULL;
 245                 }
 246                 write_unlock(&em_tree->lock);
 247
 248                 /* once for us */
 249                 free_extent_map(em);
 250                 /* once for the tree*/
 251                 free_extent_map(em);
 252         }
 253         if (split)
 254                 free_extent_map(split);
 255         if (split2)
 256                 free_extent_map(split2);
 257         return 0;
 258 }
 259
 260 /*
 261  * this is very complex, but the basic idea is to drop all extents
 262  * in the range start - end.  hint_block is filled in with a block number
 263  * that would be a good hint to the block allocator for this file.
 264  *
 265  * If an extent intersects the range but is not entirely inside the range
 266  * it is either truncated or split.  Anything entirely inside the range
 267  * is deleted from the tree.
 268  */
 269 int btrfs_drop_extents(struct btrfs_trans_handle *trans, struct inode *inode,
 270                        u64 start, u64 end, u64 *hint_byte, int drop_cache)
 271 {
 272         struct btrfs_root *root = BTRFS_I(inode)->root;
 273         struct extent_buffer *leaf;
 274         struct btrfs_file_extent_item *fi;
 275         struct btrfs_path *path;
 276         struct btrfs_key key;
 277         struct btrfs_key new_key;
 278         u64 search_start = start;
 279         u64 disk_bytenr = 0;
 280         u64 num_bytes = 0;
 281         u64 extent_offset = 0;
 282         u64 extent_end = 0;
 283         int del_nr = 0;
 284         int del_slot = 0;
 285         int extent_type;
 286         int recow;
 287         int ret;
 288
 289         if (drop_cache)
 290                 btrfs_drop_extent_cache(inode, start, end - 1, 0);
 291
 292         path = btrfs_alloc_path();
 293         if (!path)
 294                 return -ENOMEM;
 295
 296         while (1) {
 297                 recow = 0;
 298                 ret = btrfs_lookup_file_extent(trans, root, path, inode->i_ino,
 299                                                search_start, -1);
 300                 if (ret < 0)
 301                         break;
 302                 if (ret > 0 && path->slots[0] > 0 && search_start == start) {
 303                         leaf = path->nodes[0];
 304                         btrfs_item_key_to_cpu(leaf, &key, path->slots[0] - 1);
 305                         if (key.objectid == inode->i_ino &&
 306                             key.type == BTRFS_EXTENT_DATA_KEY)
 307                                 path->slots[0]--;
 308                 }
 309                 ret = 0;
 310 next_slot:
 311                 leaf = path->nodes[0];
 312                 if (path->slots[0] >= btrfs_header_nritems(leaf)) {
 313                         BUG_ON(del_nr > 0);
 314                         ret = btrfs_next_leaf(root, path);
 315                         if (ret < 0)
 316                                 break;
 317                         if (ret > 0) {
 318                                 ret = 0;
 319                                 break;
 320                         }
 321                         leaf = path->nodes[0];
 322                         recow = 1;
 323                 }
 324
 325                 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 326                 if (key.objectid > inode->i_ino ||
 327                     key.type > BTRFS_EXTENT_DATA_KEY || key.offset >= end)
 328                         break;
 329
 330                 fi = btrfs_item_ptr(leaf, path->slots[0],
 331                                     struct btrfs_file_extent_item);
 332                 extent_type = btrfs_file_extent_type(leaf, fi);
 333
 334                 if (extent_type == BTRFS_FILE_EXTENT_REG ||
 335                     extent_type == BTRFS_FILE_EXTENT_PREALLOC) {
 336                         disk_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 337                         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 338                         extent_offset = btrfs_file_extent_offset(leaf, fi);
 339                         extent_end = key.offset +
 340                                 btrfs_file_extent_num_bytes(leaf, fi);
 341                 } else if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 342                         extent_end = key.offset +
 343                                 btrfs_file_extent_inline_len(leaf, fi);
 344                 } else {
 345                         WARN_ON(1);
 346                         extent_end = search_start;
 347                 }
 348
 349                 if (extent_end <= search_start) {
 350                         path->slots[0]++;
 351                         goto next_slot;
 352                 }
 353
 354                 search_start = max(key.offset, start);
 355                 if (recow) {
 356                         btrfs_release_path(root, path);
 357                         continue;
 358                 }
 359
 360                 /*
 361                  *     | - range to drop - |
 362                  *  | -------- extent -------- |
 363                  */
 364                 if (start > key.offset && end < extent_end) {
 365                         BUG_ON(del_nr > 0);
 366                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 367
 368                         memcpy(&new_key, &key, sizeof(new_key));
 369                         new_key.offset = start;
 370                         ret = btrfs_duplicate_item(trans, root, path,
 371                                                    &new_key);
 372                         if (ret == -EAGAIN) {
 373                                 btrfs_release_path(root, path);
 374                                 continue;
 375                         }
 376                         if (ret < 0)
 377                                 break;
 378
 379                         leaf = path->nodes[0];
 380                         fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 381                                             struct btrfs_file_extent_item);
 382                         btrfs_set_file_extent_num_bytes(leaf, fi,
 383                                                         start - key.offset);
 384
 385                         fi = btrfs_item_ptr(leaf, path->slots[0],
 386                                             struct btrfs_file_extent_item);
 387
 388                         extent_offset += start - key.offset;
 389                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 390                         btrfs_set_file_extent_num_bytes(leaf, fi,
 391                                                         extent_end - start);
 392                         btrfs_mark_buffer_dirty(leaf);
 393
 394                         if (disk_bytenr > 0) {
 395                                 ret = btrfs_inc_extent_ref(trans, root,
 396                                                 disk_bytenr, num_bytes, 0,
 397                                                 root->root_key.objectid,
 398                                                 new_key.objectid,
 399                                                 start - extent_offset);
 400                                 BUG_ON(ret);
 401                                 *hint_byte = disk_bytenr;
 402                         }
 403                         key.offset = start;
 404                 }
 405                 /*
 406                  *  | ---- range to drop ----- |
 407                  *      | -------- extent -------- |
 408                  */
 409                 if (start <= key.offset && end < extent_end) {
 410                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 411
 412                         memcpy(&new_key, &key, sizeof(new_key));
 413                         new_key.offset = end;
 414                         btrfs_set_item_key_safe(trans, root, path, &new_key);
 415
 416                         extent_offset += end - key.offset;
 417                         btrfs_set_file_extent_offset(leaf, fi, extent_offset);
 418                         btrfs_set_file_extent_num_bytes(leaf, fi,
 419                                                         extent_end - end);
 420                         btrfs_mark_buffer_dirty(leaf);
 421                         if (disk_bytenr > 0) {
 422                                 inode_sub_bytes(inode, end - key.offset);
 423                                 *hint_byte = disk_bytenr;
 424                         }
 425                         break;
 426                 }
 427
 428                 search_start = extent_end;
 429                 /*
 430                  *       | ---- range to drop ----- |
 431                  *  | -------- extent -------- |
 432                  */
 433                 if (start > key.offset && end >= extent_end) {
 434                         BUG_ON(del_nr > 0);
 435                         BUG_ON(extent_type == BTRFS_FILE_EXTENT_INLINE);
 436
 437                         btrfs_set_file_extent_num_bytes(leaf, fi,
 438                                                         start - key.offset);
 439                         btrfs_mark_buffer_dirty(leaf);
 440                         if (disk_bytenr > 0) {
 441                                 inode_sub_bytes(inode, extent_end - start);
 442                                 *hint_byte = disk_bytenr;
 443                         }
 444                         if (end == extent_end)
 445                                 break;
 446
 447                         path->slots[0]++;
 448                         goto next_slot;
 449                 }
 450
 451                 /*
 452                  *  | ---- range to drop ----- |
 453                  *    | ------ extent ------ |
 454                  */
 455                 if (start <= key.offset && end >= extent_end) {
 456                         if (del_nr == 0) {
 457                                 del_slot = path->slots[0];
 458                                 del_nr = 1;
 459                         } else {
 460                                 BUG_ON(del_slot + del_nr != path->slots[0]);
 461                                 del_nr++;
 462                         }
 463
 464                         if (extent_type == BTRFS_FILE_EXTENT_INLINE) {
 465                                 inode_sub_bytes(inode,
 466                                                 extent_end - key.offset);
 467                                 extent_end = ALIGN(extent_end,
 468                                                    root->sectorsize);
 469                         } else if (disk_bytenr > 0) {
 470                                 ret = btrfs_free_extent(trans, root,
 471                                                 disk_bytenr, num_bytes, 0,
 472                                                 root->root_key.objectid,
 473                                                 key.objectid, key.offset -
 474                                                 extent_offset);
 475                                 BUG_ON(ret);
 476                                 inode_sub_bytes(inode,
 477                                                 extent_end - key.offset);
 478                                 *hint_byte = disk_bytenr;
 479                         }
 480
 481                         if (end == extent_end)
 482                                 break;
 483
 484                         if (path->slots[0] + 1 < btrfs_header_nritems(leaf)) {
 485                                 path->slots[0]++;
 486                                 goto next_slot;
 487                         }
 488
 489                         ret = btrfs_del_items(trans, root, path, del_slot,
 490                                               del_nr);
 491                         BUG_ON(ret);
 492
 493                         del_nr = 0;
 494                         del_slot = 0;
 495
 496                         btrfs_release_path(root, path);
 497                         continue;
 498                 }
 499
 500                 BUG_ON(1);
 501         }
 502
 503         if (del_nr > 0) {
 504                 ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 505                 BUG_ON(ret);
 506         }
 507
 508         btrfs_free_path(path);
 509         return ret;
 510 }
 511
 512 static int extent_mergeable(struct extent_buffer *leaf, int slot,
 513                             u64 objectid, u64 bytenr, u64 *start, u64 *end)
 514 {
 515         struct btrfs_file_extent_item *fi;
 516         struct btrfs_key key;
 517         u64 extent_end;
 518
 519         if (slot < 0 || slot >= btrfs_header_nritems(leaf))
 520                 return 0;
 521
 522         btrfs_item_key_to_cpu(leaf, &key, slot);
 523         if (key.objectid != objectid || key.type != BTRFS_EXTENT_DATA_KEY)
 524                 return 0;
 525
 526         fi = btrfs_item_ptr(leaf, slot, struct btrfs_file_extent_item);
 527         if (btrfs_file_extent_type(leaf, fi) != BTRFS_FILE_EXTENT_REG ||
 528             btrfs_file_extent_disk_bytenr(leaf, fi) != bytenr ||
 529             btrfs_file_extent_compression(leaf, fi) ||
 530             btrfs_file_extent_encryption(leaf, fi) ||
 531             btrfs_file_extent_other_encoding(leaf, fi))
 532                 return 0;
 533
 534         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 535         if ((*start && *start != key.offset) || (*end && *end != extent_end))
 536                 return 0;
 537
 538         *start = key.offset;
 539         *end = extent_end;
 540         return 1;
 541 }
 542
 543 /*
 544  * Mark extent in the range start - end as written.
 545  *
 546  * This changes extent type from 'pre-allocated' to 'regular'. If only
 547  * part of extent is marked as written, the extent will be split into
 548  * two or three.
 549  */
 550 int btrfs_mark_extent_written(struct btrfs_trans_handle *trans,
 551                               struct inode *inode, u64 start, u64 end)
 552 {
 553         struct btrfs_root *root = BTRFS_I(inode)->root;
 554         struct extent_buffer *leaf;
 555         struct btrfs_path *path;
 556         struct btrfs_file_extent_item *fi;
 557         struct btrfs_key key;
 558         struct btrfs_key new_key;
 559         u64 bytenr;
 560         u64 num_bytes;
 561         u64 extent_end;
 562         u64 orig_offset;
 563         u64 other_start;
 564         u64 other_end;
 565         u64 split;
 566         int del_nr = 0;
 567         int del_slot = 0;
 568         int ret;
 569
 570         btrfs_drop_extent_cache(inode, start, end - 1, 0);
 571
 572         path = btrfs_alloc_path();
 573         BUG_ON(!path);
 574 again:
 575         split = start;
 576         key.objectid = inode->i_ino;
 577         key.type = BTRFS_EXTENT_DATA_KEY;
 578         key.offset = split;
 579
 580         ret = btrfs_search_slot(trans, root, &key, path, -1, 1);
 581         if (ret > 0 && path->slots[0] > 0)
 582                 path->slots[0]--;
 583
 584         leaf = path->nodes[0];
 585         btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
 586         BUG_ON(key.objectid != inode->i_ino ||
 587                key.type != BTRFS_EXTENT_DATA_KEY);
 588         fi = btrfs_item_ptr(leaf, path->slots[0],
 589                             struct btrfs_file_extent_item);
 590         BUG_ON(btrfs_file_extent_type(leaf, fi) !=
 591                BTRFS_FILE_EXTENT_PREALLOC);
 592         extent_end = key.offset + btrfs_file_extent_num_bytes(leaf, fi);
 593         BUG_ON(key.offset > start || extent_end < end);
 594
 595         bytenr = btrfs_file_extent_disk_bytenr(leaf, fi);
 596         num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi);
 597         orig_offset = key.offset - btrfs_file_extent_offset(leaf, fi);
 598
 599         while (start > key.offset || end < extent_end) {
 600                 if (key.offset == start)
 601                         split = end;
 602
 603                 memcpy(&new_key, &key, sizeof(new_key));
 604                 new_key.offset = split;
 605                 ret = btrfs_duplicate_item(trans, root, path, &new_key);
 606                 if (ret == -EAGAIN) {
 607                         btrfs_release_path(root, path);
 608                         goto again;
 609                 }
 610                 BUG_ON(ret < 0);
 611
 612                 leaf = path->nodes[0];
 613                 fi = btrfs_item_ptr(leaf, path->slots[0] - 1,
 614                                     struct btrfs_file_extent_item);
 615                 btrfs_set_file_extent_num_bytes(leaf, fi,
 616                                                 split - key.offset);
 617
 618                 fi = btrfs_item_ptr(leaf, path->slots[0],
 619                                     struct btrfs_file_extent_item);
 620
 621                 btrfs_set_file_extent_offset(leaf, fi, split - orig_offset);
 622                 btrfs_set_file_extent_num_bytes(leaf, fi,
 623                                                 extent_end - split);
 624                 btrfs_mark_buffer_dirty(leaf);
 625
 626                 ret = btrfs_inc_extent_ref(trans, root, bytenr, num_bytes, 0,
 627                                            root->root_key.objectid,
 628                                            inode->i_ino, orig_offset);
 629                 BUG_ON(ret);
 630
 631                 if (split == start) {
 632                         key.offset = start;
 633                 } else {
 634                         BUG_ON(start != key.offset);
 635                         path->slots[0]--;
 636                         extent_end = end;
 637                 }
 638         }
 639
 640         fi = btrfs_item_ptr(leaf, path->slots[0],
 641                             struct btrfs_file_extent_item);
 642
 643         other_start = end;
 644         other_end = 0;
 645         if (extent_mergeable(leaf, path->slots[0] + 1, inode->i_ino,
 646                              bytenr, &other_start, &other_end)) {
 647                 extent_end = other_end;
 648                 del_slot = path->slots[0] + 1;
 649                 del_nr++;
 650                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 651                                         0, root->root_key.objectid,
 652                                         inode->i_ino, orig_offset);
 653                 BUG_ON(ret);
 654         }
 655         other_start = 0;
 656         other_end = start;
 657         if (extent_mergeable(leaf, path->slots[0] - 1, inode->i_ino,
 658                              bytenr, &other_start, &other_end)) {
 659                 key.offset = other_start;
 660                 del_slot = path->slots[0];
 661                 del_nr++;
 662                 ret = btrfs_free_extent(trans, root, bytenr, num_bytes,
 663                                         0, root->root_key.objectid,
 664                                         inode->i_ino, orig_offset);
 665                 BUG_ON(ret);
 666         }
 667         if (del_nr == 0) {
 668                 btrfs_set_file_extent_type(leaf, fi,
 669                                            BTRFS_FILE_EXTENT_REG);
 670                 btrfs_mark_buffer_dirty(leaf);
 671                 goto out;
 672         }
 673
 674         fi = btrfs_item_ptr(leaf, del_slot - 1,
 675                             struct btrfs_file_extent_item);
 676         btrfs_set_file_extent_type(leaf, fi, BTRFS_FILE_EXTENT_REG);
 677         btrfs_set_file_extent_num_bytes(leaf, fi,
 678                                         extent_end - key.offset);
 679         btrfs_mark_buffer_dirty(leaf);
 680
 681         ret = btrfs_del_items(trans, root, path, del_slot, del_nr);
 682         BUG_ON(ret);
 683 out:
 684         btrfs_free_path(path);
 685         return 0;
 686 }
 687
 688 /*
 689  * this gets pages into the page cache and locks them down, it also properly
 690  * waits for data=ordered extents to finish before allowing the pages to be
 691  * modified.
 692  */
 693 static noinline int prepare_pages(struct btrfs_root *root, struct file *file,
 694                          struct page **pages, size_t num_pages,
 695                          loff_t pos, unsigned long first_index,
 696                          unsigned long last_index, size_t write_bytes)
 697 {
 698         int i;
 699         unsigned long index = pos >> PAGE_CACHE_SHIFT;
 700         struct inode *inode = fdentry(file)->d_inode;
 701         int err = 0;
 702         u64 start_pos;
 703         u64 last_pos;
 704
 705         start_pos = pos & ~((u64)root->sectorsize - 1);
 706         last_pos = ((u64)index + num_pages) << PAGE_CACHE_SHIFT;
 707
 708         if (start_pos > inode->i_size) {
 709                 err = btrfs_cont_expand(inode, start_pos);
 710                 if (err)
 711                         return err;
 712         }
 713
 714         memset(pages, 0, num_pages * sizeof(struct page *));
 715 again:
 716         for (i = 0; i < num_pages; i++) {
 717                 pages[i] = grab_cache_page(inode->i_mapping, index + i);
 718                 if (!pages[i]) {
 719                         err = -ENOMEM;
 720                         BUG_ON(1);
 721                 }
 722                 wait_on_page_writeback(pages[i]);
 723         }
 724         if (start_pos < inode->i_size) {
 725                 struct btrfs_ordered_extent *ordered;
 726                 lock_extent(&BTRFS_I(inode)->io_tree,
 727                             start_pos, last_pos - 1, GFP_NOFS);
 728                 ordered = btrfs_lookup_first_ordered_extent(inode,
 729                                                             last_pos - 1);
 730                 if (ordered &&
 731                     ordered->file_offset + ordered->len > start_pos &&
 732                     ordered->file_offset < last_pos) {
 733                         btrfs_put_ordered_extent(ordered);
 734                         unlock_extent(&BTRFS_I(inode)->io_tree,
 735                                       start_pos, last_pos - 1, GFP_NOFS);
 736                         for (i = 0; i < num_pages; i++) {
 737                                 unlock_page(pages[i]);
 738                                 page_cache_release(pages[i]);
 739                         }
 740                         btrfs_wait_ordered_range(inode, start_pos,
 741                                                  last_pos - start_pos);
 742                         goto again;
 743                 }
 744                 if (ordered)
 745                         btrfs_put_ordered_extent(ordered);
 746
 747                 clear_extent_bits(&BTRFS_I(inode)->io_tree, start_pos,
 748                                   last_pos - 1, EXTENT_DIRTY | EXTENT_DELALLOC |
 749                                   EXTENT_DO_ACCOUNTING,
 750                                   GFP_NOFS);
 751                 unlock_extent(&BTRFS_I(inode)->io_tree,
 752                               start_pos, last_pos - 1, GFP_NOFS);
 753         }
 754         for (i = 0; i < num_pages; i++) {
 755                 clear_page_dirty_for_io(pages[i]);
 756                 set_page_extent_mapped(pages[i]);
 757                 WARN_ON(!PageLocked(pages[i]));
 758         }
 759         return 0;
 760 }
 761
 762 static ssize_t btrfs_file_write(struct file *file, const char __user *buf,
 763                                 size_t count, loff_t *ppos)
 764 {
 765         loff_t pos;
 766         loff_t start_pos;
 767         ssize_t num_written = 0;
 768         ssize_t err = 0;
 769         int ret = 0;
 770         struct inode *inode = fdentry(file)->d_inode;
 771         struct btrfs_root *root = BTRFS_I(inode)->root;
 772         struct page **pages = NULL;
 773         int nrptrs;
 774         struct page *pinned[2];
 775         unsigned long first_index;
 776         unsigned long last_index;
 777         int will_write;
 778
 779         will_write = ((file->f_flags & O_SYNC) || IS_SYNC(inode) ||
 780                       (file->f_flags & O_DIRECT));
 781
 782         nrptrs = min((count + PAGE_CACHE_SIZE - 1) / PAGE_CACHE_SIZE,
 783                      PAGE_CACHE_SIZE / (sizeof(struct page *)));
 784         pinned[0] = NULL;
 785         pinned[1] = NULL;
 786
 787         pos = *ppos;
 788         start_pos = pos;
 789
 790         vfs_check_frozen(inode->i_sb, SB_FREEZE_WRITE);
 791
 792         /* do the reserve before the mutex lock in case we have to do some
 793          * flushing.  We wouldn't deadlock, but this is more polite.
 794          */
 795         err = btrfs_reserve_metadata_for_delalloc(root, inode, 1);
 796         if (err)
 797                 goto out_nolock;
 798
 799         mutex_lock(&inode->i_mutex);
 800
 801         current->backing_dev_info = inode->i_mapping->backing_dev_info;
 802         err = generic_write_checks(file, &pos, &count, S_ISBLK(inode->i_mode));
 803         if (err)
 804                 goto out;
 805
 806         if (count == 0)
 807                 goto out;
 808
 809         err = file_remove_suid(file);
 810         if (err)
 811                 goto out;
 812
 813         file_update_time(file);
 814
 815         pages = kmalloc(nrptrs * sizeof(struct page *), GFP_KERNEL);
 816
 817         /* generic_write_checks can change our pos */
 818         start_pos = pos;
 819
 820         BTRFS_I(inode)->sequence++;
 821         first_index = pos >> PAGE_CACHE_SHIFT;
 822         last_index = (pos + count) >> PAGE_CACHE_SHIFT;
 823
 824         /*
 825          * there are lots of better ways to do this, but this code
 826          * makes sure the first and last page in the file range are
 827          * up to date and ready for cow
 828          */
 829         if ((pos & (PAGE_CACHE_SIZE - 1))) {
 830                 pinned[0] = grab_cache_page(inode->i_mapping, first_index);
 831                 if (!PageUptodate(pinned[0])) {
 832                         ret = btrfs_readpage(NULL, pinned[0]);
 833                         BUG_ON(ret);
 834                         wait_on_page_locked(pinned[0]);
 835                 } else {
 836                         unlock_page(pinned[0]);
 837                 }
 838         }
 839         if ((pos + count) & (PAGE_CACHE_SIZE - 1)) {
 840                 pinned[1] = grab_cache_page(inode->i_mapping, last_index);
 841                 if (!PageUptodate(pinned[1])) {
 842                         ret = btrfs_readpage(NULL, pinned[1]);
 843                         BUG_ON(ret);
 844                         wait_on_page_locked(pinned[1]);
 845                 } else {
 846                         unlock_page(pinned[1]);
 847                 }
 848         }
 849
 850         while (count > 0) {
 851                 size_t offset = pos & (PAGE_CACHE_SIZE - 1);
 852                 size_t write_bytes = min(count, nrptrs *
 853                                         (size_t)PAGE_CACHE_SIZE -
 854                                          offset);
 855                 size_t num_pages = (write_bytes + PAGE_CACHE_SIZE - 1) >>
 856                                         PAGE_CACHE_SHIFT;
 857
 858                 WARN_ON(num_pages > nrptrs);
 859                 memset(pages, 0, sizeof(struct page *) * nrptrs);
 860
 861                 ret = btrfs_check_data_free_space(root, inode, write_bytes);
 862                 if (ret)
 863                         goto out;
 864
 865                 ret = prepare_pages(root, file, pages, num_pages,
 866                                     pos, first_index, last_index,
 867                                     write_bytes);
 868                 if (ret) {
 869                         btrfs_free_reserved_data_space(root, inode,
 870                                                        write_bytes);
 871                         goto out;
 872                 }
 873
 874                 ret = btrfs_copy_from_user(pos, num_pages,
 875                                            write_bytes, pages, buf);
 876                 if (ret) {
 877                         btrfs_free_reserved_data_space(root, inode,
 878                                                        write_bytes);
 879                         btrfs_drop_pages(pages, num_pages);
 880                         goto out;
 881                 }
 882
 883                 ret = dirty_and_release_pages(NULL, root, file, pages,
 884                                               num_pages, pos, write_bytes);
 885                 btrfs_drop_pages(pages, num_pages);
 886                 if (ret) {
 887                         btrfs_free_reserved_data_space(root, inode,
 888                                                        write_bytes);
 889                         goto out;
 890                 }
 891
 892                 if (will_write) {
 893                         filemap_fdatawrite_range(inode->i_mapping, pos,
 894                                                  pos + write_bytes - 1);
 895                 } else {
 896                         balance_dirty_pages_ratelimited_nr(inode->i_mapping,
 897                                                            num_pages);
 898                         if (num_pages <
 899                             (root->leafsize >> PAGE_CACHE_SHIFT) + 1)
 900                                 btrfs_btree_balance_dirty(root, 1);
 901                         btrfs_throttle(root);
 902                 }
 903
 904                 buf += write_bytes;
 905                 count -= write_bytes;
 906                 pos += write_bytes;
 907                 num_written += write_bytes;
 908
 909                 cond_resched();
 910         }
 911 out:
 912         mutex_unlock(&inode->i_mutex);
 913         if (ret)
 914                 err = ret;
 915         btrfs_unreserve_metadata_for_delalloc(root, inode, 1);
 916
 917 out_nolock:
 918         kfree(pages);
 919         if (pinned[0])
 920                 page_cache_release(pinned[0]);
 921         if (pinned[1])
 922                 page_cache_release(pinned[1]);
 923         *ppos = pos;
 924
 925         /*
 926          * we want to make sure fsync finds this change
 927          * but we haven't joined a transaction running right now.
 928          *
 929          * Later on, someone is sure to update the inode and get the
 930          * real transid recorded.
 931          *
 932          * We set last_trans now to the fs_info generation + 1,
 933          * this will either be one more than the running transaction
 934          * or the generation used for the next transaction if there isn't
 935          * one running right now.
 936          */
 937         BTRFS_I(inode)->last_trans = root->fs_info->generation + 1;
 938
 939         if (num_written > 0 && will_write) {
 940                 struct btrfs_trans_handle *trans;
 941
 942                 err = btrfs_wait_ordered_range(inode, start_pos, num_written);
 943                 if (err)
 944                         num_written = err;
 945
 946                 if ((file->f_flags & O_SYNC) || IS_SYNC(inode)) {
 947                         trans = btrfs_start_transaction(root, 1);
 948                         ret = btrfs_log_dentry_safe(trans, root,
 949                                                     file->f_dentry);
 950                         if (ret == 0) {
 951                                 ret = btrfs_sync_log(trans, root);
 952                                 if (ret == 0)
 953                                         btrfs_end_transaction(trans, root);
 954                                 else
 955                                         btrfs_commit_transaction(trans, root);
 956                         } else if (ret != BTRFS_NO_LOG_SYNC) {
 957                                 btrfs_commit_transaction(trans, root);
 958                         } else {
 959                                 btrfs_end_transaction(trans, root);
 960                         }
 961                 }
 962                 if (file->f_flags & O_DIRECT) {
 963                         invalidate_mapping_pages(inode->i_mapping,
 964                               start_pos >> PAGE_CACHE_SHIFT,
 965                              (start_pos + num_written - 1) >> PAGE_CACHE_SHIFT);
 966                 }
 967         }
 968         current->backing_dev_info = NULL;
 969         return num_written ? num_written : err;
 970 }
 971
 972 int btrfs_release_file(struct inode *inode, struct file *filp)
 973 {
 974         /*
 975          * ordered_data_close is set by settattr when we are about to truncate
 976          * a file from a non-zero size to a zero size.  This tries to
 977          * flush down new bytes that may have been written if the
 978          * application were using truncate to replace a file in place.
 979          */
 980         if (BTRFS_I(inode)->ordered_data_close) {
 981                 BTRFS_I(inode)->ordered_data_close = 0;
 982                 btrfs_add_ordered_operation(NULL, BTRFS_I(inode)->root, inode);
 983                 if (inode->i_size > BTRFS_ORDERED_OPERATIONS_FLUSH_LIMIT)
 984                         filemap_flush(inode->i_mapping);
 985         }
 986         if (filp->private_data)
 987                 btrfs_ioctl_trans_end(filp);
 988         return 0;
 989 }
 990
 991 /*
 992  * fsync call for both files and directories.  This logs the inode into
 993  * the tree log instead of forcing full commits whenever possible.
 994  *
 995  * It needs to call filemap_fdatawait so that all ordered extent updates are
 996  * in the metadata btree are up to date for copying to the log.
 997  *
 998  * It drops the inode mutex before doing the tree log commit.  This is an
 999  * important optimization for directories because holding the mutex prevents
1000  * new operations on the dir while we write to disk.
1001  */
1002 int btrfs_sync_file(struct file *file, struct dentry *dentry, int datasync)
1003 {
1004         struct inode *inode = dentry->d_inode;
1005         struct btrfs_root *root = BTRFS_I(inode)->root;
1006         int ret = 0;
1007         struct btrfs_trans_handle *trans;
1008
1009
1010         /* we wait first, since the writeback may change the inode */
1011         root->log_batch++;
1012         /* the VFS called filemap_fdatawrite for us */
1013         btrfs_wait_ordered_range(inode, 0, (u64)-1);
1014         root->log_batch++;
1015
1016         /*
1017          * check the transaction that last modified this inode
1018          * and see if its already been committed
1019          */
1020         if (!BTRFS_I(inode)->last_trans)
1021                 goto out;
1022
1023         /*
1024          * if the last transaction that changed this file was before
1025          * the current transaction, we can bail out now without any
1026          * syncing
1027          */
1028         mutex_lock(&root->fs_info->trans_mutex);
1029         if (BTRFS_I(inode)->last_trans <=
1030             root->fs_info->last_trans_committed) {
1031                 BTRFS_I(inode)->last_trans = 0;
1032                 mutex_unlock(&root->fs_info->trans_mutex);
1033                 goto out;
1034         }
1035         mutex_unlock(&root->fs_info->trans_mutex);
1036
1037         /*
1038          * ok we haven't committed the transaction yet, lets do a commit
1039          */
1040         if (file && file->private_data)
1041                 btrfs_ioctl_trans_end(file);
1042
1043         trans = btrfs_start_transaction(root, 1);
1044         if (!trans) {
1045                 ret = -ENOMEM;
1046                 goto out;
1047         }
1048
1049         ret = btrfs_log_dentry_safe(trans, root, dentry);
1050         if (ret < 0)
1051                 goto out;
1052
1053         /* we've logged all the items and now have a consistent
1054          * version of the file in the log.  It is possible that
1055          * someone will come in and modify the file, but that's
1056          * fine because the log is consistent on disk, and we
1057          * have references to all of the file's extents
1058          *
1059          * It is possible that someone will come in and log the
1060          * file again, but that will end up using the synchronization
1061          * inside btrfs_sync_log to keep things safe.
1062          */
1063         mutex_unlock(&dentry->d_inode->i_mutex);
1064
1065         if (ret != BTRFS_NO_LOG_SYNC) {
1066                 if (ret > 0) {
1067                         ret = btrfs_commit_transaction(trans, root);
1068                 } else {
1069                         ret = btrfs_sync_log(trans, root);
1070                         if (ret == 0)
1071                                 ret = btrfs_end_transaction(trans, root);
1072                         else
1073                                 ret = btrfs_commit_transaction(trans, root);
1074                 }
1075         } else {
1076                 ret = btrfs_end_transaction(trans, root);
1077         }
1078         mutex_lock(&dentry->d_inode->i_mutex);
1079 out:
1080         return ret > 0 ? EIO : ret;
1081 }
1082
1083 static const struct vm_operations_struct btrfs_file_vm_ops = {
1084         .fault          = filemap_fault,
1085         .page_mkwrite   = btrfs_page_mkwrite,
1086 };
1087
1088 static int btrfs_file_mmap(struct file  *filp, struct vm_area_struct *vma)
1089 {
1090         vma->vm_ops = &btrfs_file_vm_ops;
1091         file_accessed(filp);
1092         return 0;
1093 }
1094
1095 const struct file_operations btrfs_file_operations = {
1096         .llseek         = generic_file_llseek,
1097         .read           = do_sync_read,
1098         .aio_read       = generic_file_aio_read,
1099         .splice_read    = generic_file_splice_read,
1100         .write          = btrfs_file_write,
1101         .mmap           = btrfs_file_mmap,
1102         .open           = generic_file_open,
1103         .release        = btrfs_release_file,
1104         .fsync          = btrfs_sync_file,
1105         .unlocked_ioctl = btrfs_ioctl,
1106 #ifdef CONFIG_COMPAT
1107         .compat_ioctl   = btrfs_ioctl,
1108 #endif
1109 };